app.py

import os
import torch
import torch.nn as nn
import timm
from torchvision import transforms
from torchvision.transforms.functional import InterpolationMode
import gradio as gr
from PIL import Image
import torch.nn.functional as F

# 전역 설정
CFG = {
    'IMG_SIZE': 224
}

class MultiLabelClassificationModel(nn.Module):
    def __init__(self, num_labels):
        super(MultiLabelClassificationModel, self).__init__()

        # 이미지 특징 추출
        self.cnn = timm.create_model("timm/convnext_base.clip_laion2b_augreg_ft_in12k_in1k", pretrained=True, drop_rate=0.05, drop_path_rate=0.05, in_chans=3)

        # 멀티 라벨 분류 헤드
        self.classification_head = nn.Linear(1000, num_labels)

    def forward(self, images):
        # CNN
        features = self.cnn(images)
        features_flat = features.view(features.size(0), -1)

        # 멀티 라벨 분류
        logits = self.classification_head(features_flat)
        # probs = torch.sigmoid(logits)

        return logits

test_transform = transforms.Compose([
        transforms.Resize(size=(CFG['IMG_SIZE'], CFG['IMG_SIZE']), interpolation=InterpolationMode.BICUBIC),
        transforms.ToTensor(),
        transforms.Normalize(mean = [0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225]),
        ])

model = MultiLabelClassificationModel(num_labels=13)
model.load_state_dict(torch.load(f'checkpoint.tar')['model_state_dict'])
model.eval()  # 모델을 평가 모드로 설정

# 미리 설정한 라벨 목록
labels = ['Mold', 'blight', 'greening', 'healthy', 'measles',
           'mildew', 'mite', 'rot', 'rust', 'scab', 'scorch', 'spot', 'virus']

def predict(image_path):
    image = Image.open(image_path)
    image = test_transform(image).unsqueeze(0)
    with torch.no_grad():
        logits = model(image)
    probs = F.softmax(logits, dim=1)  # softmax를 적용하여 확률 값으로 변환
    result = {label: float(probs[0][i]) for i, label in enumerate(labels)}
    return result

app = gr.Interface(
    fn=predict,
    inputs=gr.Image(type='filepath'),
    outputs=gr.Label(),
    title='Multi-Label Image Classification',
    description='Automatically classify images into the following categories: ' + ', '.join(labels) + '.'
)

app.launch(share=True)